Recent advances in precision medicine for individualized immunosuppression

Abstract

Purpose of Review:

The current tools to proactively guide and individualize immunosuppression in solid organ transplantation are limited. Despite continued improvements in post-transplant outcomes, the side effects of over- or under-immunosuppression are common. This review is intended to highlight recent advances in individualized immunosuppression.

Recent Findings:

There has been a great focus on genomic information to predict drug dose requirements, specifically on single nucleotide polymorphisms of CYP3A5 and ABCB1. Furthermore, biomarker studies have developed ways to better predict clinical outcomes, such as graft rejection.

Summary:

The integration of advanced computing tools, such as artificial neural networks and machine learning, with genome sequencing has led to intriguing findings on individual or group-specific dosing requirements. Rapid computing allows for processing of data and discovering otherwise undetected clinical patterns. Genetic polymorphisms of CYP3A5 and ABCB1 have yielded results to suggest varying dose requirements correlated with race and sex. Newly proposed biomarkers offer precise and noninvasive ways to monitor patient’s status. Cell-free DNA quantitation is increasingly explored as an indicator of allograft injury and rejection, which can help avoid unneeded biopsies and more frequently monitor graft function.

Introduction

Transplantation is one of the greatest achievements of modern medicine. Its success has depended in large part to advances in modulating the immune system to allay the risks of allograft rejection, while simultaneously minimizing the risks of infection and carcinogenesis, as well as other side effects. In the absence of a clinically validated measure of level of immunosuppression, therapeutic drug monitoring is the standard-of-care, specifically measuring blood levels of calcineurin or mTOR inhibitors. Tacrolimus is the most commonly used calcineurin inhibitor for solid organ transplantation. It has a narrow therapeutic window and wide pharmacokinetic variability, with approximately 25% variation in bioavailability between individuals [1]. In immunosuppression management, tacrolimus is frequently used in combination with steroids, antiproliferatives, and occasionally mTOR inhibitors [2]. Dosing guidelines and treatment protocols vary among centers, but the current standard-of-care dosing practice is physician-guided, with close monitoring of tacrolimus whole-blood trough concentrations. The high intraindividual variability of tacrolimus often leads to inaccurate dosing [3*]. Many factors interfere with the pharmacokinetics of tacrolimus, including drug-drug interactions, body weight, age, race/ethnicity, and sex [4–6]. Toxic side-effects of tacrolimus dosing are common, including serious clinical events such as neurotoxicity, nephrotoxicity, cardiomyopathy, graft injury, and death [7]. The current accepted method is “one system fits all,” but this is not precise enough to account for individual variations. Thus, there is an urgent need for more systematic methods and strategies in immunosuppression dosing.

In the past few years, there has been much innovation and interest in personalized medicine and patient-centered care, which has prompted many international large-scale immunosuppression personalization studies [8*]. Here, we review recent advances in precision and personalized medicine for transplant patients. We will discuss recently formulated models based on genomic information and other factors for the prediction of risk factors and dose requirements of immunosuppression after transplantation.

Pharmacogenomics

Genetic polymorphisms of CYP3A5 are predictors of drug dosing requirements. Different alleles in the CYP3A5 gene contribute to variable tacrolimus bioavailability, the most studied being CYP3A5*1 and *3 [9*]. CYP3A enzymes are highly involved in the metabolism of tacrolimus, influencing its clearance and bioavailability [10*]. A large proportion of Caucasians are homozygous for CYP3A5*3 [11]. Moreover, CYP3A5*6 and *7 genes are predominantly present in African Americans [12*]. This variation allows for new opportunities for genotype-guided dosing.

Genotype-guided Dosing

A large-scale genotyping study was conducted with kidney transplant patients to formulate an African American-specific tacrolimus clearance model [13**]. This was conducted with the knowledge of genotypic allelic differences between African Americans and Caucasians that contribute to differences in tacrolimus metabolism, a generally higher tacrolimus dose requirement for African Americans, and worse outcomes after transplantation in African American patients. CYP3A5*3, *6, and *7 were present in 50% of the sample population, which reduced tacrolimus clearance by 15%. A synergistic effect of 47% decrease in tacrolimus clearance was seen if two of the alleles (CYP3A5*3/*3, *3/*6, *3/*7, *6/*7, or *6/*6) were carried by an individual. The CYP3A5*6 and *7 alleles were not found in Caucasian patients.

Machine Learning

Machine learning (ML) exploits the processing power of computer systems to analyze complex data and patterns [14]. ML has been used to predict risk factors, treatment outcomes, and dosing [15,16]. This extends a tool to fuel the growing interest in personalized medicine. In a 2017 prospective study by Tang et al., eight different machine learning models were compared with multiple linear regression, the current method of genotype-based tacrolimus dosing [17**]. 80% of recruited subjects were randomly selected to generate ML algorithms (“derivation cohort”), while the remaining were used to test the final algorithm. The machine learning models were constructed using demographic, clinical and genetic factors. Demographic factors included ethnicity, age, and gender. Clinical values of concomitant medications (i.e. omeprazole, metoprolol, furosemide, and calcium channel blockers), hypertension, diabetes, concomitant medication, anemia, cardiac insufficiency, hepatic and renal dysfunction were noted. Single nucleotide polymorphisms of CYP3A5, CYP3A4 and ABCB1 were analyzed in the model; specifically, more focus was given to CYP3A5 *3 genotypes. In this large cohort of 1045 renal transplant patients, 50.6% carried the CYP3A5*3 GG genotype in the derivation cohort and 46.9% were carriers in the validation cohort. The regression tree ML method outperformed multiple linear regression by 4%. However, worth noting is that random forest regression ML model had the smallest mean absolute error, and the regression tree algorithm performed much better in intermediate tacrolimus dose ranges (2.5 – 4 mg/day).

Artificial Neural Networks

Artificial neural networks (ANNs) have also been explored as a potential tool for personalized dosing [18,19]. ANNs borrow the fundamentals of machine learning and use the human neural infrastructure for learning and pattern-seeking processes [20]. Neurons are arranged in neuronal layers – typically input, hidden, and output layers – that are parallel to each other and feed information into the consecutive layer [21]. Recent applications of ANNs include insulin, radiation dosing, and remifentanil dosing [18,22,23].

One prospective study of 129 renal transplant patients, excluding those with impaired liver function and combined organ transplantation tested ANNs in tacrolimus dosing [24**]. These patients received a combined immunosuppression regimen of tacrolimus, prednisolone, and mycophenolate mofetil. Tacrolimus doses were adjusted based on trough levels. Various clinical outcomes were measured within the 6-month study period. This study used the basic setup of a 3-layer ANN for tacrolimus dosing. In the input layer, factors of creatinine, BMI, age, gender, ABCB1 polymorphisms, and CYP3A5*3 genotypes were considered to generate a bioavailability value (ratio of plasma concentration with oral dose of tacrolimus). Conclusions about the bioavailability of tacrolimus were made using the ANN. Men with higher bioavailability of tacrolimus had mutant alleles in CYP3A5 A>G and ABCB1 3435C>T, whereas men with lower bioavailability had mutant alleles in ABCB1 1236C>T and 2677G>T/A. Women with higher bioavailability carried a mutant allele in CYP3A5 A>G, and women with lower bioavailability showed mutant alleles in ABCB1 3435C>T, 1236C>T, and 2677G>T/A. Researchers found a synergistic effect with the combined genotype of CYP3A5*3 and ABCB1 1236TT/2677GG that yielded the highest bioavailability of tacrolimus. Moreover, ABCB1 2677 G>T is found to increase the risk factor for post-renal transplant diabetes mellitus. Generally, CYP3A5 A>G SNP increased bioavailability of tacrolimus, while ABCB1 1236C>T decreased it. The predicted bioavailability of tacrolimus had a great overlap with experimental levels. Gene-gene interactions between CYP3A5*3, ABCB1 2677, and ABCB1 1236 further contribute to the genotypic effects on tacrolimus bioavailability.

Mathematical Approach

Despite beginning many decades ago, the use of mathematical approaches to building an iterative dosing model is only slowly increasing in use more recently, mostly in radiation dosing [25]. A recent effort composed a differential mathematical model for the mechanism of immune response and BK viral infection after renal transplant [26*]. This was an attempt to find a balance between over-immunosuppression and under-immunosuppression after renal transplant. Over-suppression creates a compromised immune system that paves the way for viral infections post-transplant, such as BK virus that leads to nephropathy. BK viral plasma load and creatinine levels were taken from a renal transplant patient diagnosed with BK viremia within the first three months after transplant. Variables of creatinine, infected cells, susceptible cells, free BK virus, allo-specific CD8+T-cells targeting kidneys, CBK virus-specific CD8+T-cells were considered in the generation of this mathematical dosing model. The current model does portray the biological process behind BK viral infection and graft loss; however, the precision needs to be improved in further studies for any possible future clinical adoption.

Phenomapping

Phenomapping uses genotypic information to predict the phenotypic response or phenomic outcome. In many pharmaceutical applications, the clinical manifestations of biological factors are analyzed to find relationships between them for use in precision medicine [27]. In a 2018 longitudinal study, Bakir et al. correlated clinical outcomes after cardiac transplant with immunosuppression and a mass collection of biomarker, genetic, and laboratory data [28*]. Analysis showed that a faster decrease in immunosuppression doses correlated with poor clinical outcomes, such as cardiac allograft vasculopathy, myocardial infarction, graft loss, rejection, death, and/or retransplantation. They found that combination drug therapies for immunosuppression, specifically tacrolimus and mycophenolate, made no differences in clinical outcomes. Moreover, poor clinical endpoints were associated with increased class I and II human leukocyte antigen antibody titers, increased expression of FLT3 genes (through analysis of FLT3 network), and downregulation of March8 and WDNR40A genes. Similar attempts to associate phenotypes with endpoint clinical outcomes have also been previously attempted [29].

Phenotypic Personalized Medicine

Given the complexity of the genetic, epigenetic, environmental, and myriad of other factors that can all affect biological responses to perturbation, a mechanism-independent, phenotype-based approach has been employed to optimize drug dosing [30,31]. A parabolic personalized dosing (PPD) method guided by tacrolimus trough levels was created in 2016 by Zarrinpar et al. Tacrolimus doses and trough levels of three consecutive days were used to generate a second order approximation of the dose-response curve [32]. This parabolic surface was then used to guide subsequent doses to bring or keep the trough levels into the target range (e.g. 6–8 ng/ml). Recalibration of the PPD surface is performed with regimen changes (e.g. going on or off hemodialysis, steroid taper, antibiotic use, or other new drug administration) using a similar approach for the generation of a new phenotypic response surface. This method was tested prospectively in a randomized controlled trial with eight liver transplant patients. Physician-guided dosing was used for the first 10 days after transplant. The patients were then randomized into 2 groups: control and PPD. The control group remained on physician-guided tacrolimus dosing, whereas the test group was dosed using the PPD method. PPD patients were less frequently out of targeted trough range (54.2 ± 4.3%) and had less variability overall (50–60%), in comparison with the control group with 72.6 ± 14.3% out of range and 61–90% variability. Though this study only adjusted the dose of one immunosuppressant (i.e. tacrolimus), two or more drug combination doses can very well be predicted in a similar manner to generate a multi-dimensional PPD.

Biomarkers

Recent immunologic and pharmacogenetic noninvasive biomarker studies are prognostic of clinical outcomes and allow for better patient monitoring after transplantation [33*]. Tolerance to kidney transplantation correlates with the increase in B cell specific genes IGKV1D-13 and IGLL-1 [34*]. A recent prospective, multicenter study with post-renal transplant patients related biomarkers to graft function [35*]. At a single given time from study entry to 2 years after transplant, 25–30% of patients displayed IGKV1D-13 and IGKV4–1 genes that are representative of tolerance. There was a reduction to 13.7% of patients displaying the two genes at any time. Immunosuppression regimens also influenced biomarker values. Specifically, those who received tacrolimus and not mycophenolate mofetil and corticosteroids saw increases in the expression of these biomarkers. Patients who were predicted to be tolerant using these biomarkers demonstrated promising increases in eGFR values and reductions in serum creatinine levels.

In another study, Suthanthiran et al. proposed a three gene-signature predictor for acute cellular rejection (ACR) through expression profiles of 18S-normalized CD3ε mRNA, IP-10 mRNA, and 18S rRNA extracted from urine samples of post-renal transplant patients [36]. Urine samples were taken from 485 patients during various times from three days to one year after transplant. Patients showed elevated levels of the three markers at an average of 20 days before biopsy verified rejection. This prognostic method for predicting and verifying ACR in renal transplant patients has high specificity and sensitivity of >79%. In 2018, the use of this predictor has been patented [43]

Specific miRNAa have also been associated with acute graft rejection in liver transplant patients [37*]. Hepatic miR-301a was shown to be involved in the mechanism of IL-6 synthesis. This finding suggests that perhaps miR-301a can be a biomarker for a more precise and objective method to defining acute rejection in liver transplant patients. Other advancements have been made to use biomarkers as a prognosis or diagnosis tool for rejection and other post-transplant outcomes.

Levels of donor-derived cell-free DNA (dd-cfDNA) in plasma have been developed recently as a safe and reliable assessment of allograft injury and rejection. A prospective study of 102 kidney transplant patients compared dd-cfDNA with 107 biopsy reports [38*]. dd-cfDNA levels were able to discriminate between antibody-mediated rejection (ABMR), T-cell mediated rejection (TCMR), and no rejection with high precision. Specifically, the discrimination between ABMR and non-ABMR status was especially strong using dd-cfDNA values; dd-cfDNA values are more elevated in ABMR (median of 2.9%) than TCMR. Moreover, type IB TCMR (median of 1.2%) was associated with higher dd-cfDNA levels than in type IA TCMR (median of 0.2%). dd-cfDNA presents a plausible alternative to biopsies for patients on anticoagulation, as well as a routine graft surveillance measure.

A recent study improved the detection of dd-cfDNA in renal transplant patients without knowledge of donor genotypes using SNP-based mmPCR-NGS (multiplexed PCR followed by next generation sequencing) in 217 biopsy-matched samples [39**]. Compared with the standard eGFR monitoring, dd-cfDNA was able to distinguish between acute rejection and non-rejection with high specificity (72.6%) and sensitivity (88.7%). Specifically, patients with acute rejection had much higher dd-cfDNA levels (median 2.32%) than patients without rejection (median 0.47%), which included borderline rejection, stable allografts, and other injuries.

Liver damage could be quantified in various ways, with the current standard as measuring serum aspartate aminotransferase (AST) and alanine aminotransferase (ALT) levels. However, the use of AST/ALT is not a precise method of distinguishing the source of injury [40,41]. Here too, groups have used cfDNA released from dying hepatocytes to aid in diagnosis [42**]. Tissue-specificity is achieved by characterizing hepatic specific DNA methylation patterns. ITIH4, IGF2R, and VTN genes are selected as loci, in which further analysis showed that sequences in these loci had hepatocyte-specific unmethylated sites (CpG sites). Serum samples were obtained from patients who had liver transplant, partial hepatectomy, sepsis, and Duchenne muscular dystrophy. In these patients, high cfDNA levels were noted right after surgical procedures and levels declined with time. Transplant patients had elevated cfDNA that corresponded to biopsy confirmed rejection, and half displayed elevated levels prior to detectable rejection. In sepsis, with cell death in multiple tissue types, patients saw cfDNA levels correlating with AST and ALT. Duchenne muscular dystrophy patients typically have high AST and ALT from non-liver origins. In these patients, hepatocyte cfDNA levels were in the normal range, which confirms the hepatocyte specificity of cfDNA. These can be used to titrate immunosuppression and apply computational techniques to optimize and individualize these combinations.

Conclusions

Recent advances from the past two years in precision methods of immunosuppression after transplantation have focused on the discoveries of different methods to predict certain clinical outcomes, such as graft injury and infection. However, with one exception models that prospectively guide immunosuppression dosing have been lacking [32]. Recently there have been many promising biomarkers, but some factors to consider are cost-effectiveness and practicality in the clinical setting. More importantly, knowledge about biomarkers need to be used to quantitate immunosuppression level, in order to adjust drugs and drug combinations for enhanced clinical performance. There is an unmet need for precision medicine protocols to yield personalized immunosuppression dosing regimens.

Key Points.

• Despite the importance of practice- and center-specific immunosuppression protocols, transplant recipients need individualized immunosuppression based on their unique clinical scenarios; to do so, transplant physicians currently use a combination of empirical assessments of the immune status of each recipient to make clinical immunosuppression decisions.

• Multiple studies show potential applications of precision medicine in aiding physicians to assess the immune status of patients and then to individualize immunosuppression.

• Most published studies thus far have focused on prediction of and correlation with clinical outcomes, rather than to prospectively guide immunosuppression therapy.

• A lack of mature precision dosing protocols applicable in clinical settings calls for future developments and research.